Rank Level Fusion of Multimodal Biometrics Using Particle Swarm Optimization

Multimodal biometric system combines information from multiple biometric modalities to uniquely identify an individual. Therefore, it overcomes the limitations of unimodal biometric system, such as inter-class similarity, non-universality and susceptibility to circumvention. One approach of fusing information for multimodal biometrics is rank level fusion. It aggregates the rank lists from individual biometric matchers into a single rank list. In this paper, rank level fusion is formulated as an optimization problem. In this context, a particle swarm optimization (PSO) based approach is proposed to fuse rank lists from several biometric modalities. It minimizes the summation of distances of an aggregated list with each input list for individual biometric modalities. Weighted Spearman footrule distance metric is used for estimating the said distance between a pair of rank lists. Superiority of the proposed particle swarm optimization based method over several other rank level fusion methods is demonstrated experimentally using two multimodal biometric datasets. Similarly, the proposed particle swarm optimization based method also exhibits superior performance than existing score level fusion methods.